Machine learning system for transaction reconciliation

Patent metadata

Field	Value
Publication number	US-11416867-B2
Application number	US-202016906044-A
Country	US
Kind code	B2
Filing date	Jun 19, 2020
Priority date	May 6, 2020
Publication date	Aug 16, 2022
Grant date	Aug 16, 2022

Abstract

A device may receive transaction data associated with transactions. The transaction data may be associated with transaction entries that are associated with the transactions. The device may process, using a matching model, the transaction entries to classify the transaction entries into a set of matched transaction entries and a set of unmatched transaction entries. The device may update a transaction grouping model based on the set of matched transaction entries to create an updated transaction grouping model. The device may determine, using the updated transaction grouping model, that a subset of the set of unmatched transaction entries are associated with a same transaction. The device may classify the subset of the set of unmatched transaction entries as grouped transaction entries. The device may provide an indication that the grouped transaction entries and the set of matched transaction entries are reconciled transactions.

First claim

What is claimed is:

1. A method, comprising:
receiving, by a device, transaction data that is associated with transaction entries,
wherein the transaction entries identify information associated with a plurality of transactions;

classifying, by the device and using a machine learning model, the transaction entries into a set of matched transaction entries and a set of unmatched transaction entries;
combining, by the device, a first matched transaction entry, of the set of matched transaction entries, and a second matched transaction entry, of the set of matched transaction entries, into a first transaction pair;
labeling, by the device, the first transaction pair to indicate that the first matched transaction entry and the second matched transaction entry belong to a same transaction of the plurality of transactions;
combining, by the device, a third matched transaction entry, of the set of matched transaction entries, and a fourth matched transaction entry, of the set of matched transaction entries, into a second transaction pair;
labeling, by the device, the second transaction pair to indicate that the third matched transaction entry and the fourth matched transaction entry belong to a different transaction of the plurality of transactions;
identifying, by the device, a feature that is associated with a subset of matched transaction entries, of the set of matched transaction entries, that are associated with the same transaction, of the plurality of transactions;
generating, by the device, an updated machine learning model,
wherein the updated machine learning model is generated using feature information associated with the feature,
wherein the updated machine learning model is generated based on labeling the first transaction pair to indicate that the first matched transaction entry and the second matched transaction entry belong to the same transaction of the plurality of transactions,
wherein the updated machine learning model is generated based on labeling the second transaction pair to indicate that the third matched transaction entry and the fourth matched transaction entry belong to the different transaction of the plurality of transactions,
wherein the updated machine learning model is trained based on historical transaction data associated with a plurality of previous transactions;

grouping, by the device and using the updated machine learning model, the set of unmatched transaction entries based on probabilities that individual unmatched transaction entries are associated with the same transaction;
classifying, by the device, grouped transaction entries, of the set of unmatched transaction entries, and the set of matched transaction entries as reconciled transactions;
determining, by the device, that an ungrouped transaction entry, of the set of unmatched transaction entries, is associated with missing transaction data;
displaying, by the device and via a user interface, a request for confirmation of the reconciled transactions;
receiving, by the device and via the user interface, an input confirming the reconciled transactions or an input rejecting the reconciled transactions; and
selectively:
updating, by the device and based on receiving the input confirming the reconciled transactions, a data structure to mark the transactions as reconciled and to include predicted transaction data, and
prompting, by the device and based on receiving the input rejecting the reconciled transactions via the user interface, a user to input information that identifies the missing transaction data.

2. The method of claim 1 , further comprising:
labeling the set of matched transaction entries with match identifiers that are associated with corresponding transactions of the plurality of transactions; and
wherein identifying the feature that is associated with the subset of matched transaction entries that are associated with the same transaction comprises:
identifying the subset of matched transaction entries based on a match identifier of the match identifiers.

3. The method of claim 1 , wherein identifying the feature that is associated with the subset of matched transaction entries that are associated with the same transaction comprises:
combining transaction information, associated with pairs of transaction entries of the set of matched transaction entries, into sets of combined transaction information; and
generating a co-occurrence matrix based on the sets of combined transaction information; and
wherein generating the updated machine learning model comprises:
generating the updated machine learning model based on the co-occurrence matrix.

4. The method of claim 1 , wherein generating the updated machine learning model comprises:
training the updated machine learning model based on the feature information; and
wherein the method further comprises:
identifying, based on training the updated machine learning model based on the feature information, the feature in the set of unmatched transaction entries or a subsequently identified set of unmatched transaction entries.

5. The method of claim 1 , wherein determining that the ungrouped transaction entry, of the set of unmatched transaction entries, is associated with the missing transaction data comprises:
determining, using an autoencoder, a first error metric associated with the grouped transaction entries;
determining, using the autoencoder or a different autoencoder, a second error metric associated with the set of matched transaction entries;
comparing the first error metric and the second error metric; and
determining that the ungrouped transaction entry is associated with the missing transaction data based on comparing the first error metric and the second error metric.

6. The method of claim 1 , further comprising:
determining a characteristic associated with the missing transaction data; and
outputting information that identifies the characteristic.

7. The method of claim 1 , further comprising:
indicating that the plurality of transactions include:
the reconciled transactions,
an unreconciled transaction that is associated with the ungrouped transaction entry, and
the missing transaction data.

8. A device, comprising:
one or more memories; and
one or more processors coupled to the one or more memories, configured to:
receive transaction data associated with a plurality of transactions,
wherein the transaction data is associated with transaction entries that are associated with the plurality of transactions;

classify, using a machine learning model, the transaction entries into a set of matched transaction entries and a set of unmatched transaction entries;
combine a first matched transaction entry, of the set of matched transaction entries, and a second matched transaction entry, of the set of matched transaction entries, into a first transaction pair;
label the first transaction pair to indicate that the first matched transaction entry and the second matched transaction entry belong to a same transaction of the plurality of transactions;
combine a third matched transaction entry, of the set of matched transaction entries, and a fourth matched transaction entry, of the set of matched transaction entries, into a second transaction pair;
label the second transaction pair to indicate that the third matched transaction entry and the fourth matched transaction entry belong to a different transaction of the plurality of transactions;
generate an updated machine learning model based on the set of matched transaction entries,
wherein the updated machine learning model is generated based on labeling the first transaction pair to indicate that the first matched transaction entry and the second matched transaction entry belong to the

Assignees

• Accenture Global Solutions Ltd

Inventors

• Lopes Maria
• Raval Contractor Gopali
• Lakshminarayanan Subhashini
• S R Prasanna
• Rane Parag
• Deo Arati
• Shahzad Azimullah
• Sriramula Vallinath

Classifications

• G06N5/01

  Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06N3/044

  Recurrent networks, e.g. Hopfield networks · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/0442

  characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU] · CPC title